PV Arc Fault Diagnosis and Modeling Methods: State of the Art and Perspectives

The global photovoltaic (PV) power capacity is growing exponentially. However, the undetected arc faults would pose a severe fire hazard to PV systems, so various advanced diagnosis techniques have been proposed especially in the last few years. This talk presents a comprehensive review of state-of-the-art techniques for arc fault diagnosis and modeling methods in PV systems, and the development trend of future diagnosis methods is also discussed. Diagnosis methods viewed from physical and electrical signals of PV arc faults have been proposed for a few decades. Their capabilities and limitations are discussed, compared, and summarized in detail. By acquiring electromagnetic radiation and sound characteristics of arc faults, diagnosis methods based on physical signals have the advantage of the accurate identification. However, these methods show limitations for large-scale PV systems due to the increasing interference factors in the exposed environment. Through signal processing methods such as time-domain methods, fast Fourier transform and time-frequency transforms, much more works focus on diagnosis methods based on electrical signals. Recently, diagnosis methods with good switching noise and system transition immunity have been introduced. For instance, the existing Db9-based features would cause nuisance trip for the arc fault detection in grid-connected PV systems. The Rbio3.1-based features are proposed to achieve better arc fault recognition ability. Since the field testing is costly and time consuming, precisely modeling arc faults becomes more critical. Different types of arc fault models including dynamical state model, stationary state model, and high-frequency component model have been reviewed and compared. In addition, future trends about PV arc fault diagnosis methods are outlined. It is predicted that facing more complex arc fault conditions, the data processing chip development and machine learning based classifier are of great significance to improve the detection accuracy of diagnosis methods. Also, the detection reliability of diagnosis methods would be significantly improved without increasing the computation time significantly.